Real-time tracking of highly articulated structures in the presence of noisy measurements
|
|
- Justin Jacobs
- 5 years ago
- Views:
Transcription
1 Real-time tracking of highly articulated structures in the presence of noisy measurements T. Drummond R. Cipolla Department of Engineering University of Cambridge Cambridge, UK CB2 1PZ Department of Engineering University of Cambridge Cambridge, UK CB2 1PZ Abstract This paper presents a novel approach for model-based realtime tracking of highly articulated structures such as humans. This approach is based on an algorithm which efficiently propagates statistics of probability distributions through a kinematic chain to obtain maximum a posteriori estimates of the motion of the entire structure. This algorithm yields the least squares solution in linear time (in the number of components of the model) and can also be applied to non-gaussian statistics using a simple but powerful trick. The resulting implementation runs in real-time on standard hardware without any pre-processing of the video data and can thus operate on live video. Results from experiments performed using this system are presented and discussed. 1. Introduction Visual tracking of complex, highly articulated structures is an important technology for several domains. In particular there has been much interest in tracking human motion [9]. There are many applications including tasks such as surveillance, motion capture and human-computer interaction. The problem addressed by this paper is that of realtime tracking an articulated structure in the view of one or more cameras. The system is model-based [16, 12, 7, 4] and uses a CAD model that comprises piecewise rigid components with curved surfaces and known kinematic constraints. Several tasks such as human-computer interaction require real-time performance. Attaining this is difficult and there are relatively few examples of such systems. An exception is Pfinder [17] which does so by limiting its processing to tracking coloured blobs in the image plane. Black and Jepson [1] also use an image-based approach with multiple eigenspaces to recognise hand gestures. Cham and Rehg [3] suggest that real-time performance is not possible in three dimensions and use a two-dimensional scaled prismatic model. Hel-Or and Werman [8] use a fully articulated three-dimensional model but use an O(N 3 ) algorithm to obtain a least squares solution for the pose. Here we present an O(N) algorithm which propagates statistics of probability distributions along the kinematic chain to obtain the maximum a posteriori solution for the pose in real-time. Where these statistics are Gaussian, the same least squares solution is obtained. Further the algorithm can be adapted to operate iteratively with robust (non-gaussian) statistics. Delamarre and Faugeras [4] also use an O(N) algorithm and give timings which are essentially real-time, but make use of a powerful (but comparatively expensive) pre-processing technique (Geodesic active contours) which yields impressive results. In this work, we represent the pose of each component of the model separately as an element of the group of rigid body motions in three dimensions, SE(3). This is a redundant representation and requires that the articulation constraints be explicitly represented, however it has the advantages that it provides a symmetric representation and is not reliant upon accurate localisation of some key component to identify the pose of the remainder of the model. This is by contrast to [13, 2] which use a minimal parameterisation based on a tree structure. We make use of the exponential map [2, 15] connecting the Lie algebra of SE(3) with the group to represent motions, and use the adjoint representation of the group to transform quantities in the algebra between different coordinate frames. To obtain robust real-time performance in the presence of noisy measurements, it is important to use a strong statistical framework. MacCormick and Blake [14] use a refinement of the Condensation algorithm [11] which partitions the search space in order to reduce computational complexity. This was extended by Deutscher et. al. [5] who used coupled monte-carlo techniques to track highly articulated human motion although the computational cost was prohibitive for real-time application. Cham and Rehg [3] also use a particle based method, but capture the structure of local modes of the posterior distribution in each particle and thus need fewer particles. The approach employed here is to 1
2 use an iterative re-weighted least-squares technique which allows the use of the fast O(N) algorithm with non-gaussian statistics. For applications that deal with existing imagery, typically only a single view is available and several systems have been developed which can handle this [3, 2, 13, 1]. The system described here can also operate using a single view, although the performance is greatly improved PSfrag when replacements more views are available. We use an edge-based approach (as [7]). In our case the model is rendered first and matching edges are then sought in the image rather than processing the image to detect edges first and then matching those to the model. Currently, we only attempt to solve the motion problem (where is each component at each time step) by contrast to [12] who use data to refine the model in addition to computing pose. 2. Geometric representation 2.1. Pose The system presented here represents the pose of an articulated structure with a matrix which describes the transformation from the coordinates of each model component to each camera. These matrices form the group of Euclidean transformations on three-dimensional space, SE(3) and have the form: E = [ R ] t where R is an orthogonal 3 3 matrix with determinant 1 and t is an arbitrary 3-vector. For convenience, the bottom row of these matrices (being constant) will be omitted for the remainder of the paper. The internal camera parameters for each camera are also stored and thus the projection matrix for each component of the model into each camera can be computed Structure In order to construct models of structures with curved surfaces, a representation based on intersections of pairs of quadrics is used. This is a convenient representation as it permits the use of standard structures such as truncated cones, cylinders and spheres as well as truncated ellipsoids and hyperbolic surfaces. These structures can be rapidly rendered by computing the normalised image conic C for each quadric Q in the view of a normalised camera with a Euclidean projection matrix E. Each quadric can be transformed into camera coordinates to give (1) Q = E T QE 1 (2) Q 1 = Q 2 = Figure 1: Q 1 is the quadric to be rendered and is clipped by the (in this case degenerate) quadric Q 2. Only the sample points on the conic image of Q 1 that correspond to points inside Q 2 are rendered (shown in bold). If the transformed quadric is written as [ ] Q A b = b T c then the conic can be computed as (3) C = Ac bb T (4) The conic C can then be rendered by mapping a unit circle (parameterised by θ) to C and generating a series of sample points (to be used in tracking, see Section 3) from a discrete set of θs. These are evenly separated with spacing p dp/dθ p by setting θ =. Each sample point generated by this procedure is then checked for visibility (in front of the camera, within the intersecting quadric and not occluded by any other pair of intersected quadrics). The articulated structures considered in this paper are constructed from these primitives. The structures comprise piecewise rigid components and each component is made up of a number of these clipped quadrics. The quadrics in each component share a common coordinate frame and are all rendered using the Euclidean projection matrix for the component. The kinematic chain is represented by specifying a parent for each component and listing the constraints that apply between the two. These are described in more detail in Section 4. Because two components may share the same parent, the kinematic chain may form a tree structure, however articulated cycles are not permitted Motion The task of this system is to compute the Euclidean matrices E for each component into each camera at each time step.
3 This problem can be reduced to computing a Euclidean motion matrix M such that E t = E t 1 M t (5) The matrices M give the transformation from the coordinate frame of a component at time t to the frame at time t 1 (coarse manual initialisation is given for the first frame). Because the motion is represented in the coordinate frame of the moving component, the motion matrices are the same for all stationary cameras viewing the scene and thus only need to be computed once per component per time frame regardless of the number of views. The matrices M can be represented by points α in the Lie algebra of SE(3) by means of the exponential map [2]. ( 6 ) M = exp α i G i (6) i=1 where G i are a set of generator matrices which form a basis for the algebra. Because the time steps are small (40ms for PAL video), the motion matrices are close to the identity and thus α should be close to the origin. The tracking task is then reduced to a six dimensional search for α for each component at each time frame. It should be noted that if a component contains just a single quadric, it will appear in the image as a single conic having just five degrees of freedom. Thus there will be (in general) a one parameter family of solutions for α for this component considered in isolation and this can only be constrained by considering the articulation constraints. 3. Statistical framework The statistical formulation used in this approach defines the probability of observing an image given a particular pose of the model in terms of the presence of edges in the image close to those rendered in the model. These edges are detected by performing a 1-dimensional search from each sample point in a direction normal to the rendered edge. These measurements are assumed independent and the probability of the image given a pose is the product of the probabilities of each measurement Gaussian statistics If these measurements were normally distributed then the maximum a posteriori pose would be given by a leastsquares solution. Such a solution can be computed very efficiently from the Jacobian of partial derivatives of each distance measurement with respect to each motion parameter α i. If the distance measurements are small, an accurate approximation to the Jacobian J can be obtained by considering the edge-normal component of the motion of the sample points with variation of α i which can be easily Figure 2: Noisy edge measurements. The solid bar indicates failure to detect an edge. computed in closed form. This can be used to obtain a maximum a posteriori estimate of the motion parameters µ for each component µ = C 1 J T m where C = [J T J] (7) and m is the vector of distance measurements. The sumsquared residual measurements S for a solution α is given by S = (α µ)c(α µ) + (m Jµ) 2. (8) This gives a probability distribution over the Lie algebra (with two arbitrary constants a and b) of the form p(α) = a exp ( b(α µ)c(α µ)) (9) 3.2. Non-Gaussian statistics In practice, the measurement distribution is found to be non- Gaussian by virtue of containing many more samples in the tails of the distribution (see Figure 2). Such distributions can be handled within the least squares framework by introducing a re-weighting function [10]. This function is evaluated for each distance measurement m in m and is used to scale m and the corresponding row of J. If the re-weighting function is w(m), then the least squares procedure iteratively converges on a solution given by the probability distribution ( m ) p(m) = exp m w(m )dm (10) 0 This approach has the advantage that distributions other than Gaussian can be modelled, convergence is fast and each iteration requires just a linear solution. The real power of this approach becomes apparent when complex systems such as kinematic chains are considered since it is relatively easy to propagate Gaussian statistics through the chain and these can then be wrapped with a re-weighting function to obtain an iterative solution to the desired statistics. The re-weighting function used in this work is w(m) = 1/(c + m ). This results in a distribution that behaves as a Gaussian for m c and a Laplacian for m c.
4 4. Highly articulated kinematic chains Pose statistics are now considered for kinematic chains. Equation (9) gives independent probability distributions for the motion parameters for each component of the model. The model is not free to move arbitrarily, however, since the articulation constraints must be respected. This means that the desired solution is one which maximises the product of the probabilities for the motion of each component and also satisfies the constraints. This can be computed efficiently (in linear time in the number of components) by propagating the statistics through the model. This is done by using the constraints present between each adjacent pair of links in the kinematic chain to obtain the maximum a posteriori motion for the entire chain. This motion will provide the least squares solution for the pose, subject to the kinematic constraints. This is achieved in two stages, shown in Algorithm 1. First the joint distribution of each component s motion and that of its parent in the kinematic chain is considered. The motion of the child component can be marginalised by allowing it to take its modal value conditional upon the motion of the parent. By doing this, the statistics of the parent component can be modified to incorporate those of the child. This process is repeated up the kinematic chain until the component at the top of the chain carries the propagated statistics for the entire chain. Second, the maximum a posteriori pose of each component is assigned, starting at the top of the chain and propagating back down the chain Propagating statistics It has been shown [6] that constraints corresponding to slides, hinges and ball joints can be linearised and are homogeneous on the values of the motion parameters of the two components α 1 and α 2 and thus have a simple form: D 12 α 1 + D 21 α 2 = 0. (11) Within α 1 -α 2 space there is an embedded manifold of motions which respect the physical articulation constraints. Equation (11) corresponds to forcing the motion to lie in the tangent space to this manifold at the origin (through which the manifold must pass). For a ball joint positioned at x, y, z in the coordinate frame of component 1, the constraint matrix D 12 is: z y D 12 = z 0 x (12) y x 0 The motion α 2 is represented in component 2 s coordinate frame which is different to that of component 1. If these coordinate frames were the same then D 21 = D 12. Thus if α 2 is transformed into component 1 s coordinate frame using the adjoint representation of the transformation between Algorithm 1 Statistical propagation on a kinematic chain for i = 1 n 1 do Let p = i s parent Marginalise α i and update µ p and C p using (21), (22) end for assign α n = µ n for i = n 1 1 do assign α i using (19) end for these coordinate frames the constraint in (11) becomes D 12 α 1 D 12 Ad(E 1 1 E 2)α 2 = 0 (13) (where E 1 and E 2 are the Euclidean projection matrices of components 1 and 2). Thus D 21 = D 12 Ad(E 1 1 E 2). (14) Given α 2 it is possible to obtain the value of α 1 which satisfies the constraints (11) and minimises the sum-squared residual (8). This can be computed by introducing Lagrange multipliers λ for the constraints and solving So 2C 1 (α 1 µ 1 ) + D T 12λ = 0 (15) giving α 1 = µ C 1 1 DT 12λ (16) D 21 α 2 + D 12 µ D 12C 1 1 DT 12λ = 0 (17) and λ = 2 [ D 12 C ] DT (D21 α 2 + D 12 µ 1 ) (18) α 1 = µ 1 C1 1 [ DT 12 D12 C1 1 1(D21 12] DT α 2 + D 12 µ 1 ) (19) The total sum-squared error for components 1 and 2 can be written as S 1,2 = (α 1 µ 1 )C 1 (α 1 µ 1 ) + (α 2 µ 2 )C 2 (α 2 µ 2 ) (20) (discarding the constant terms). By substituting the optimal value for α 1 from (19), S 1,2 can computed as a function of α 2. This gives new values for µ 2 and C 2 which take into account the optimal pose of component 1. µ 2 = µ 2 C2 1 [ DT 21 D12 C1 1 DT 12 + D 21 C2 1 ] 1 DT 21 (D 12 µ 1 + D 21 µ 2 ) (21) C 2 = C 2 + D21 T [ D12 C ] DT D21 (22) This propagates the statistics of component 1 through the articulated joint into component 2. The process can be repeated, propagating the new statistics for component 2 into the next component up the chain until the statistics for all components have been propagated into the root of the chain.
5 Figure 3: Tracking the wooden mannequin in a single view. Solid lines represent failure to find an edge on the search path At this point, the final value of µ for the root of the chain represents the maximum a posteriori value for the motion parameters α of this component, taking into account the measurements of the entire articulated structure. Equation (19) can then be used to propagate the maximum a posteriori estimate for the pose of the entire structure back down the kinematic chain so that the maximum a posteriori pose of the entire chain is obtained. As mentioned previously, this entire process is wrapped within a iterative re-weighting scheme so that the iterative behaviour of the system is equivalent to propagating the correct non-gaussian statistics along the kinematic chain. Although the use of non-gaussian statistics results in an iterative algorithm (with each iteration involving execution of Algorithm 1), this system only performs one iteration per video frame. Where the motion is large, a single iteration is found to provide a sufficiently good solution to permit tracking to continue and where it is small, the algorithm converges very rapidly Coercing the constraints Because the manifold corresponding to poses of the model which respect the articulation constraint is curved and the process outlined above is only correct to first order, the resultant pose will contain a quadratic error which violates the constraints. Thus at each stage, it is necessary to remove this error and ensure that the constraints are met exactly. Since the error is very small, it is possible to achieve this very simply by running down the kinematic chain and coercing the constraint in a non-symmetric manner. This is done by computing the logarithm of the matrix describing the transformation between each pair of components and projecting its coefficients into the right null space of the constraint matrix D 12. Find a s.t. exp ( i a ) ig i = E 1 2 E 1 (23) a = a D12 T [ D12 D12 T ] 1 D12 a (24) The Euclidean projection matrix E 1 can then be rebuilt to exactly satisfy the constraint by E1 = E2 exp ( i a i G ) i (25) 5. Experimental Results Two humanoid models have been used for experiments tracking a wooden mannequin and a human. The models are similar and both have 18 degrees of freedom. The model of the mannequin is quite accurate and fits well, while the model of the human is only approximate Mannequin An experiment was performed to track the mannequin in the view of a single camera. Because of difficulty in moving the mannequin by hand without causing catastrophic occlusion, a series of incremental movements were performed with the tracker turned off. After each movement, the tracker was operated for approximately a quarter of a second (about 6 frames) before the next movement was made. Results from this tracking sequence are shown in Figure 3. The single camera tracking system runs in real-time at PAL video frame rate (25Hz) Human For the human tracking experiment, views from three synchronised cameras were used. Single view tracking was found to be much less effective on this task due to a relatively poor fit between the human and the model. Figure 4 shows a number of frames from a tracking sequence in this experiment. 6. Summary and Conclusions A novel method for fast tracking of highly complex articulated structures has been presented. The algorithm runs in real-time at 25 Hz for a single view and 10Hz for three views. The maximum tolerable speed of motion that this algorithm can handle is limited however, particularly in the human case due to inaccuracies in the CAD model. Future work will consider methods for extending the framework to compute the structural parameters of the model.
6 References Figure 4: Tracking a human in three concurrent views (cameras distributed vertically). [1] M. J. Black and A. D. Jepson. Eigen tracking: Robust matching and tracking of articulated objects using a view based representation. In Proceedings of ECCV 96, volume 1, pages , [2] C Bregler and J Malik. Tracking people with twists and exponential maps. In Proceedings of CVPR 98, pages 8 15, [3] T-J Cham and J M Rehg. A multiple hypothesis approach to figure tracking. In Proceedings of CVPR 99, volume 2, pages , [4] Q. Delamarre and O. Faugeras. 3D articulated models and multi-view tracking with silhouttes. In Proceedings of ICCV 99, volume 2, pages , [5] J. Deutscher, A. Blake, and I. Reid. Articulated body motion capture by annealed particle filtering. In Proceedings of CVPR2000, volume 2, pages , [6] T. Drummond and R. Cipolla. Real-time tracking of multiple articulated structures in multiple views. In Proceedings of ECCV2000, volume 2, pages 20 36, [7] D M Gavrila and L S Davis. 3-d model-based tracking of humans in action: a multi-view approach. In Proceedings of CVPR 96, pages 73 80, [8] Y. Hel-Or and M. Werman. Constraint fusion for recognition and localisation of articulated objects. International Journal of Computer Vision, 19(1):5 28, [9] D. Hogg. A program to see a walking person. Image and Vision Computing, 1(1):5 20, [10] P. J. Huber. Robust Statistics. Wiley series in probability and mathematical statistics. Wiley, [11] M. Isard and A. Blake. CONDENSATION - conditional density propagation for visual tracking. International Journal of Computer Vision, 29(1):5 28, [12] I. A. Kakadiaris and D. Metaxas. Model-based estimation of 3D human motion with occlusion based on active multiviewpoint selection. In Proceedings of CVPR 1996, pages 81 87, [13] D. G. Lowe. Fitting parameterised 3-D models to images. IEEE T-PAMI, 13(5): , [14] J. MacCormick and A. Blake. Partitioned sampling, articulated objects, and interface quality hand tracking. In Proceedings of ECCV2000, volume 2, pages 3 19, [15] R.M. Murray and S.S. Sastry. A Mathematical Introduction to Robotic Manipulation. CRC Press, [16] J. M. Regh and T. Kanade. Model-based tracking of selfoccluding articulated objects. In Proceedings of ICCV 95, pages , [17] C. Wren, A. Azerbayejani, T. Darrell, and A. Pentland. Pfinder: Real-time tracking of the human body. IEEE T- PAMI, 19(7): , 1997.
Visual Tracking of Human Body with Deforming Motion and Shape Average
Visual Tracking of Human Body with Deforming Motion and Shape Average Alessandro Bissacco UCLA Computer Science Los Angeles, CA 90095 bissacco@cs.ucla.edu UCLA CSD-TR # 020046 Abstract In this work we
More information3D Model Acquisition by Tracking 2D Wireframes
3D Model Acquisition by Tracking 2D Wireframes M. Brown, T. Drummond and R. Cipolla {96mab twd20 cipolla}@eng.cam.ac.uk Department of Engineering University of Cambridge Cambridge CB2 1PZ, UK Abstract
More informationHuman Upper Body Pose Estimation in Static Images
1. Research Team Human Upper Body Pose Estimation in Static Images Project Leader: Graduate Students: Prof. Isaac Cohen, Computer Science Mun Wai Lee 2. Statement of Project Goals This goal of this project
More informationExponential Maps for Computer Vision
Exponential Maps for Computer Vision Nick Birnie School of Informatics University of Edinburgh 1 Introduction In computer vision, the exponential map is the natural generalisation of the ordinary exponential
More informationA Dynamic Human Model using Hybrid 2D-3D Representations in Hierarchical PCA Space
A Dynamic Human Model using Hybrid 2D-3D Representations in Hierarchical PCA Space Eng-Jon Ong and Shaogang Gong Department of Computer Science, Queen Mary and Westfield College, London E1 4NS, UK fongej
More informationTracking of Human Body using Multiple Predictors
Tracking of Human Body using Multiple Predictors Rui M Jesus 1, Arnaldo J Abrantes 1, and Jorge S Marques 2 1 Instituto Superior de Engenharia de Lisboa, Postfach 351-218317001, Rua Conselheiro Emído Navarro,
More informationKinematic Sets for Real-time Robust Articulated Object Tracking
Kinematic Sets for Real-time Robust Articulated Object Tracking Andrew I. Comport, Éric Marchand, François Chaumette IRISA - INRIA Rennes Campus de Beaulieu, 3542 Rennes Cedex, France. E-Mail: Firstname.Name@irisa.fr
More informationModel Based Perspective Inversion
Model Based Perspective Inversion A. D. Worrall, K. D. Baker & G. D. Sullivan Intelligent Systems Group, Department of Computer Science, University of Reading, RG6 2AX, UK. Anthony.Worrall@reading.ac.uk
More informationVisual Motion Analysis and Tracking Part II
Visual Motion Analysis and Tracking Part II David J Fleet and Allan D Jepson CIAR NCAP Summer School July 12-16, 16, 2005 Outline Optical Flow and Tracking: Optical flow estimation (robust, iterative refinement,
More informationInternational Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXIV-5/W10
BUNDLE ADJUSTMENT FOR MARKERLESS BODY TRACKING IN MONOCULAR VIDEO SEQUENCES Ali Shahrokni, Vincent Lepetit, Pascal Fua Computer Vision Lab, Swiss Federal Institute of Technology (EPFL) ali.shahrokni,vincent.lepetit,pascal.fua@epfl.ch
More informationArticulated Body Motion Capture by Annealed Particle Filtering
Articulated Body Motion Capture by Annealed Particle Filtering Jonathan Deutscher University of Oxford Dept. of Engineering Science Oxford, O13PJ United Kingdom jdeutsch@robots.ox.ac.uk Andrew Blake Microsoft
More informationObject Tracking with an Adaptive Color-Based Particle Filter
Object Tracking with an Adaptive Color-Based Particle Filter Katja Nummiaro 1, Esther Koller-Meier 2, and Luc Van Gool 1,2 1 Katholieke Universiteit Leuven, ESAT/VISICS, Belgium {knummiar,vangool}@esat.kuleuven.ac.be
More informationModel-Based Silhouette Extraction for Accurate People Tracking
Model-Based Silhouette Extraction for Accurate People Tracking Ralf Plaenkers and Pascal Fua Computer Graphics Lab (LIG) Computer Graphics Lab, EPFL, CH-1015 Lausanne, Switzerland In European Conference
More informationInferring 3D Body Pose from Silhouettes using Activity Manifold Learning
CVPR 4 Inferring 3D Body Pose from Silhouettes using Activity Manifold Learning Ahmed Elgammal and Chan-Su Lee Department of Computer Science, Rutgers University, New Brunswick, NJ, USA {elgammal,chansu}@cs.rutgers.edu
More informationPerception and Action using Multilinear Forms
Perception and Action using Multilinear Forms Anders Heyden, Gunnar Sparr, Kalle Åström Dept of Mathematics, Lund University Box 118, S-221 00 Lund, Sweden email: {heyden,gunnar,kalle}@maths.lth.se Abstract
More informationCS 565 Computer Vision. Nazar Khan PUCIT Lectures 15 and 16: Optic Flow
CS 565 Computer Vision Nazar Khan PUCIT Lectures 15 and 16: Optic Flow Introduction Basic Problem given: image sequence f(x, y, z), where (x, y) specifies the location and z denotes time wanted: displacement
More information3D Human Motion Analysis and Manifolds
D E P A R T M E N T O F C O M P U T E R S C I E N C E U N I V E R S I T Y O F C O P E N H A G E N 3D Human Motion Analysis and Manifolds Kim Steenstrup Pedersen DIKU Image group and E-Science center Motivation
More informationA New Hierarchical Particle Filtering for Markerless Human Motion Capture
A New Hierarchical Particle Filtering for Markerless Human Motion Capture Yuanqiang Dong, Guilherme N. DeSouza Electrical and Computer Engineering Department University of Missouri, Columbia, MO, USA Abstract
More informationDiscrete approximations to continuous curves
Proceedings of the 6 IEEE International Conference on Robotics and Automation Orlando, Florida - May 6 Discrete approximations to continuous curves Sean B. Andersson Department of Aerospace and Mechanical
More informationHierarchical Part-Based Human Body Pose Estimation
Hierarchical Part-Based Human Body Pose Estimation R. Navaratnam A. Thayananthan P. H. S. Torr R. Cipolla University of Cambridge Oxford Brookes Univeristy Department of Engineering Cambridge, CB2 1PZ,
More informationEigenTracking: Robust Matching and Tracking of Articulated Objects Using a View-Based Representation
EigenTracking: Robust Matching and Tracking of Articulated Objects Using a View-Based Representation Michael J. Black and Allan D. Jepson Xerox Palo Alto Research Center, 3333 Coyote Hill Road, Palo Alto,
More informationBehaviour based particle filtering for human articulated motion tracking
Loughborough University Institutional Repository Behaviour based particle filtering for human articulated motion tracking This item was submitted to Loughborough University's Institutional Repository by
More informationMarkerless Motion Capture
Markerless Motion Capture Paris Kaimakis August 31, 2004 Signal Processing Laboratory Department of Engineering University of Cambridge 2 Declaration I hereby declare that, except where specifically indicated,
More informationHand Tracking Using A Quadric Surface Model
Hand Tracking Using A Quadric Surface Model R. Cipolla B. Stenger A. Thayananthan P. H. S. Torr University of Cambridge, Department of Engineering, Trumpington Street, Cambridge, CB2 1PZ, UK Microsoft
More informationTracking Using Online Feature Selection and a Local Generative Model
Tracking Using Online Feature Selection and a Local Generative Model Thomas Woodley Bjorn Stenger Roberto Cipolla Dept. of Engineering University of Cambridge {tew32 cipolla}@eng.cam.ac.uk Computer Vision
More informationArticulated Characters
Articulated Characters Skeleton A skeleton is a framework of rigid body bones connected by articulated joints Used as an (invisible?) armature to position and orient geometry (usually surface triangles)
More informationLocal quadrics surface approximation for real-time tracking of textureless 3D rigid curved objects
Local quadrics surface approximation for real-time tracking of textureless 3D rigid curved objects Marina Atsumi Oikawa, Takafumi Taketomi, Goshiro Yamamoto Nara Institute of Science and Technology, Japan
More informationProbabilistic Tracking and Reconstruction of 3D Human Motion in Monocular Video Sequences
Probabilistic Tracking and Reconstruction of 3D Human Motion in Monocular Video Sequences Presentation of the thesis work of: Hedvig Sidenbladh, KTH Thesis opponent: Prof. Bill Freeman, MIT Thesis supervisors
More informationRobust Tracking of People by a Mobile Robotic Agent
Robust Tracking of People by a Mobile Robotic Agent Rawesak Tanawongsuwan, Alexander Stoytchev, Irfan Essa College of Computing, GVU Center, Georgia Institute of Technology Atlanta, GA 30332-0280 USA ftee
More informationVisualization and Analysis of Inverse Kinematics Algorithms Using Performance Metric Maps
Visualization and Analysis of Inverse Kinematics Algorithms Using Performance Metric Maps Oliver Cardwell, Ramakrishnan Mukundan Department of Computer Science and Software Engineering University of Canterbury
More informationChapter 9 Object Tracking an Overview
Chapter 9 Object Tracking an Overview The output of the background subtraction algorithm, described in the previous chapter, is a classification (segmentation) of pixels into foreground pixels (those belonging
More informationLayered Motion Segmentation and Depth Ordering by Tracking Edges
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. XX, NO. Y, MONTH 2003 1 Layered Motion Segmentation and Depth Ordering by Tracking Edges Paul Smith, Tom Drummond Member, IEEE, and
More informationJacobians. 6.1 Linearized Kinematics. Y: = k2( e6)
Jacobians 6.1 Linearized Kinematics In previous chapters we have seen how kinematics relates the joint angles to the position and orientation of the robot's endeffector. This means that, for a serial robot,
More informationSwitching Hypothesized Measurements: A Dynamic Model with Applications to Occlusion Adaptive Joint Tracking
Switching Hypothesized Measurements: A Dynamic Model with Applications to Occlusion Adaptive Joint Tracking Yang Wang Tele Tan Institute for Infocomm Research, Singapore {ywang, telctan}@i2r.a-star.edu.sg
More informationInverse Kinematics II and Motion Capture
Mathematical Foundations of Computer Graphics and Vision Inverse Kinematics II and Motion Capture Luca Ballan Institute of Visual Computing Comparison 0 1 A B 2 C 3 Fake exponential map Real exponential
More informationBayesian Reconstruction of 3D Human Motion from Single-Camera Video
Bayesian Reconstruction of 3D Human Motion from Single-Camera Video Nicholas R. Howe Cornell University Michael E. Leventon MIT William T. Freeman Mitsubishi Electric Research Labs Problem Background 2D
More informationCamera Calibration Using Line Correspondences
Camera Calibration Using Line Correspondences Richard I. Hartley G.E. CRD, Schenectady, NY, 12301. Ph: (518)-387-7333 Fax: (518)-387-6845 Email : hartley@crd.ge.com Abstract In this paper, a method of
More informationFusion of Multiple Tracking Algorithms for Robust People Tracking
Fusion of Multiple Tracking Algorithms for Robust People Tracking Nils T Siebel and Steve Maybank Computational Vision Group Department of Computer Science The University of Reading Reading RG6 6AY, England
More informationRobust Model-Free Tracking of Non-Rigid Shape. Abstract
Robust Model-Free Tracking of Non-Rigid Shape Lorenzo Torresani Stanford University ltorresa@cs.stanford.edu Christoph Bregler New York University chris.bregler@nyu.edu New York University CS TR2003-840
More informationEdge tracking for motion segmentation and depth ordering
Edge tracking for motion segmentation and depth ordering P. Smith, T. Drummond and R. Cipolla Department of Engineering University of Cambridge Cambridge CB2 1PZ,UK {pas1001 twd20 cipolla}@eng.cam.ac.uk
More information3D Human Tracking in a Top View Using Depth Information Recorded by the Xtion Pro-Live Camera
3D Human Tracking in a Top View Using Depth Information Recorded by the Xtion Pro-Live Camera Cyrille Migniot and Fakhreddine Ababsa IBISC - team IRA2 - University of Evry val d Essonne, France, {Cyrille.Migniot,Fakhr-Eddine.Ababsa}@ibisc.fr
More informationA novel approach to motion tracking with wearable sensors based on Probabilistic Graphical Models
A novel approach to motion tracking with wearable sensors based on Probabilistic Graphical Models Emanuele Ruffaldi Lorenzo Peppoloni Alessandro Filippeschi Carlo Alberto Avizzano 2014 IEEE International
More informationMulti-camera Tracking of Articulated Human Motion Using Motion and Shape Cues
Multi-camera Tracking of Articulated Human Motion Using Motion and Shape Cues Aravind Sundaresan and Rama Chellappa Department of Electrical and Computer Engineering, University of Maryland, College Park,
More informationTracking Multiple Pedestrians in Real-Time Using Kinematics
Abstract We present an algorithm for real-time tracking of multiple pedestrians in a dynamic scene. The algorithm is targeted for embedded systems and reduces computational and storage costs by using an
More information04 - Normal Estimation, Curves
04 - Normal Estimation, Curves Acknowledgements: Olga Sorkine-Hornung Normal Estimation Implicit Surface Reconstruction Implicit function from point clouds Need consistently oriented normals < 0 0 > 0
More informationTracking Multiple Moving Objects with a Mobile Robot
Tracking Multiple Moving Objects with a Mobile Robot Dirk Schulz 1 Wolfram Burgard 2 Dieter Fox 3 Armin B. Cremers 1 1 University of Bonn, Computer Science Department, Germany 2 University of Freiburg,
More informationFace detection in a video sequence - a temporal approach
Face detection in a video sequence - a temporal approach K. Mikolajczyk R. Choudhury C. Schmid INRIA Rhône-Alpes GRAVIR-CNRS, 655 av. de l Europe, 38330 Montbonnot, France {Krystian.Mikolajczyk,Ragini.Choudhury,Cordelia.Schmid}@inrialpes.fr
More informationComplex articulated object tracking
Author manuscript, published in "Electronic Letters on Computer Vision and Image Analysis, ELCVIA, Special issue on Articuled Motion & Deformable Objects 5, 3 (2005 20-30" Electronic Letters on Computer
More informationMultiple View Geometry in Computer Vision
Multiple View Geometry in Computer Vision Prasanna Sahoo Department of Mathematics University of Louisville 1 More on Single View Geometry Lecture 11 2 In Chapter 5 we introduced projection matrix (which
More informationUsing Subspace Constraints to Improve Feature Tracking Presented by Bryan Poling. Based on work by Bryan Poling, Gilad Lerman, and Arthur Szlam
Presented by Based on work by, Gilad Lerman, and Arthur Szlam What is Tracking? Broad Definition Tracking, or Object tracking, is a general term for following some thing through multiple frames of a video
More informationSegmentation and Tracking of Partial Planar Templates
Segmentation and Tracking of Partial Planar Templates Abdelsalam Masoud William Hoff Colorado School of Mines Colorado School of Mines Golden, CO 800 Golden, CO 800 amasoud@mines.edu whoff@mines.edu Abstract
More informationA second order algorithm for orthogonal projection onto curves and surfaces
A second order algorithm for orthogonal projection onto curves and surfaces Shi-min Hu and Johannes Wallner Dept. of Computer Science and Technology, Tsinghua University, Beijing, China shimin@tsinghua.edu.cn;
More informationModeling Adaptive Deformations during Free-form Pose Estimation
Modeling Adaptive Deformations during Free-form Pose Estimation Bodo Rosenhahn, Christian Perwass, Gerald Sommer Institut für Informatik und Praktische Mathematik Christian-Albrechts-Universität zu Kiel
More informationNonrigid Surface Modelling. and Fast Recovery. Department of Computer Science and Engineering. Committee: Prof. Leo J. Jia and Prof. K. H.
Nonrigid Surface Modelling and Fast Recovery Zhu Jianke Supervisor: Prof. Michael R. Lyu Committee: Prof. Leo J. Jia and Prof. K. H. Wong Department of Computer Science and Engineering May 11, 2007 1 2
More informationAn Adaptive Eigenshape Model
An Adaptive Eigenshape Model Adam Baumberg and David Hogg School of Computer Studies University of Leeds, Leeds LS2 9JT, U.K. amb@scs.leeds.ac.uk Abstract There has been a great deal of recent interest
More informationDynamic Time Warping for Binocular Hand Tracking and Reconstruction
Dynamic Time Warping for Binocular Hand Tracking and Reconstruction Javier Romero, Danica Kragic Ville Kyrki Antonis Argyros CAS-CVAP-CSC Dept. of Information Technology Institute of Computer Science KTH,
More informationMETRIC PLANE RECTIFICATION USING SYMMETRIC VANISHING POINTS
METRIC PLANE RECTIFICATION USING SYMMETRIC VANISHING POINTS M. Lefler, H. Hel-Or Dept. of CS, University of Haifa, Israel Y. Hel-Or School of CS, IDC, Herzliya, Israel ABSTRACT Video analysis often requires
More informationSingularity Analysis for Articulated Object Tracking
To appear in: CVPR 98, Santa Barbara, CA, June 1998 Singularity Analysis for Articulated Object Tracking Daniel D. Morris Robotics Institute Carnegie Mellon University Pittsburgh, PA 15213 James M. Rehg
More informationRobust Estimation of Human Body Kinematics from Video
Proceedings of the 1999 IEEERSJ International Conference on Intelligent Robots and Systems Robust Estimation of Human Body Kinematics from Video Ales Ude Kawato Dynamic Brain Project Exploratory Research
More informationMotion Tracking and Event Understanding in Video Sequences
Motion Tracking and Event Understanding in Video Sequences Isaac Cohen Elaine Kang, Jinman Kang Institute for Robotics and Intelligent Systems University of Southern California Los Angeles, CA Objectives!
More informationMIME: A Gesture-Driven Computer Interface
MIME: A Gesture-Driven Computer Interface Daniel Heckenberg a and Brian C. Lovell b a Department of Computer Science and Electrical Engineering, The University of Queensland, Brisbane, Australia, 4072
More informationBumptrees for Efficient Function, Constraint, and Classification Learning
umptrees for Efficient Function, Constraint, and Classification Learning Stephen M. Omohundro International Computer Science Institute 1947 Center Street, Suite 600 erkeley, California 94704 Abstract A
More informationTracking Human Body Parts Using Particle Filters Constrained by Human Biomechanics
Tracking Human Body Parts Using Particle Filters Constrained by Human Biomechanics J. Martínez 1, J.C. Nebel 2, D. Makris 2 and C. Orrite 1 1 Computer Vision Lab, University of Zaragoza, Spain 2 Digital
More informationExperiments with Edge Detection using One-dimensional Surface Fitting
Experiments with Edge Detection using One-dimensional Surface Fitting Gabor Terei, Jorge Luis Nunes e Silva Brito The Ohio State University, Department of Geodetic Science and Surveying 1958 Neil Avenue,
More informationHOUGH TRANSFORM CS 6350 C V
HOUGH TRANSFORM CS 6350 C V HOUGH TRANSFORM The problem: Given a set of points in 2-D, find if a sub-set of these points, fall on a LINE. Hough Transform One powerful global method for detecting edges
More informationStereo and Epipolar geometry
Previously Image Primitives (feature points, lines, contours) Today: Stereo and Epipolar geometry How to match primitives between two (multiple) views) Goals: 3D reconstruction, recognition Jana Kosecka
More informationSingularity Analysis of an Extensible Kinematic Architecture: Assur Class N, Order N 1
David H. Myszka e-mail: dmyszka@udayton.edu Andrew P. Murray e-mail: murray@notes.udayton.edu University of Dayton, Dayton, OH 45469 James P. Schmiedeler The Ohio State University, Columbus, OH 43210 e-mail:
More informationMarkerless Motion Capture from Monocular Videos
Markerless Motion Capture from Monocular Videos Vishal Mamania Appu Shaji Sharat Chandran Department of Computer Science &Engineering Indian Institute of Technology Bombay Mumbai, India 400 076 {vishalm,appu,sharat}@cse.iitb.ac.in
More informationAutomatic Partitioning of High Dimensional Search Spaces associated with Articulated Body Motion Capture
Automatic Partitioning of High Dimensional Search Spaces associated with Articulated Body Motion Capture Jonathan Deutscher University of Oxford Dept. of Engineering Science Oxford, OX13PJ United Kingdom
More informationExtracting Axially Symmetric Geometry From Limited 3D Range Data
Extracting Axially Symmetric Geometry From Limited 3D Range Data Andrew Willis, Xavier Orriols, Senem Velipasalar, Xavier Binefa, David B. Cooper Brown University, Providence, RI 02912 Computer Vision
More informationAn Adaptive Appearance Model Approach for Model-based Articulated Object Tracking
An Adaptive Appearance Model Approach for Model-based Articulated Object Tracking Alexandru O. Bălan Michael J. Black Department of Computer Science - Brown University Providence, RI 02912, USA {alb, black}@cs.brown.edu
More informationTracking humans from a moving platform
Tracking humans from a moving platform Larry Davis, Vasanth Philomin and Ramani Duraiswami Computer Vision Laboratory Institute for Advanced Computer Studies University of Maryland, College Park, MD 20742,
More informationCOMPUTER VISION > OPTICAL FLOW UTRECHT UNIVERSITY RONALD POPPE
COMPUTER VISION 2017-2018 > OPTICAL FLOW UTRECHT UNIVERSITY RONALD POPPE OUTLINE Optical flow Lucas-Kanade Horn-Schunck Applications of optical flow Optical flow tracking Histograms of oriented flow Assignment
More informationVisual Odometry. Features, Tracking, Essential Matrix, and RANSAC. Stephan Weiss Computer Vision Group NASA-JPL / CalTech
Visual Odometry Features, Tracking, Essential Matrix, and RANSAC Stephan Weiss Computer Vision Group NASA-JPL / CalTech Stephan.Weiss@ieee.org (c) 2013. Government sponsorship acknowledged. Outline The
More informationEvaluating Error Functions for Robust Active Appearance Models
Evaluating Error Functions for Robust Active Appearance Models Barry-John Theobald School of Computing Sciences, University of East Anglia, Norwich, UK, NR4 7TJ bjt@cmp.uea.ac.uk Iain Matthews and Simon
More informationOccluded Facial Expression Tracking
Occluded Facial Expression Tracking Hugo Mercier 1, Julien Peyras 2, and Patrice Dalle 1 1 Institut de Recherche en Informatique de Toulouse 118, route de Narbonne, F-31062 Toulouse Cedex 9 2 Dipartimento
More informationMultiple View Geometry in Computer Vision Second Edition
Multiple View Geometry in Computer Vision Second Edition Richard Hartley Australian National University, Canberra, Australia Andrew Zisserman University of Oxford, UK CAMBRIDGE UNIVERSITY PRESS Contents
More informationInverse Kinematics (part 1) CSE169: Computer Animation Instructor: Steve Rotenberg UCSD, Winter 2018
Inverse Kinematics (part 1) CSE169: Computer Animation Instructor: Steve Rotenberg UCSD, Winter 2018 Welman, 1993 Inverse Kinematics and Geometric Constraints for Articulated Figure Manipulation, Chris
More informationExploring Curve Fitting for Fingers in Egocentric Images
Exploring Curve Fitting for Fingers in Egocentric Images Akanksha Saran Robotics Institute, Carnegie Mellon University 16-811: Math Fundamentals for Robotics Final Project Report Email: asaran@andrew.cmu.edu
More informationParticle Filtering. CS6240 Multimedia Analysis. Leow Wee Kheng. Department of Computer Science School of Computing National University of Singapore
Particle Filtering CS6240 Multimedia Analysis Leow Wee Kheng Department of Computer Science School of Computing National University of Singapore (CS6240) Particle Filtering 1 / 28 Introduction Introduction
More informationCamera calibration. Robotic vision. Ville Kyrki
Camera calibration Robotic vision 19.1.2017 Where are we? Images, imaging Image enhancement Feature extraction and matching Image-based tracking Camera models and calibration Pose estimation Motion analysis
More informationMathematics of a Multiple Omni-Directional System
Mathematics of a Multiple Omni-Directional System A. Torii A. Sugimoto A. Imiya, School of Science and National Institute of Institute of Media and Technology, Informatics, Information Technology, Chiba
More information3D Human Body Tracking using Deterministic Temporal Motion Models
3D Human Body Tracking using Deterministic Temporal Motion Models Raquel Urtasun and Pascal Fua Computer Vision Laboratory EPFL CH-1015 Lausanne, Switzerland raquel.urtasun, pascal.fua@epfl.ch Abstract.
More informationOcclusion Robust Multi-Camera Face Tracking
Occlusion Robust Multi-Camera Face Tracking Josh Harguess, Changbo Hu, J. K. Aggarwal Computer & Vision Research Center / Department of ECE The University of Texas at Austin harguess@utexas.edu, changbo.hu@gmail.com,
More informationSimultaneous Pose and Correspondence Determination using Line Features
Simultaneous Pose and Correspondence Determination using Line Features Philip David, Daniel DeMenthon, Ramani Duraiswami, and Hanan Samet Department of Computer Science, University of Maryland, College
More informationHuman Body Model Acquisition and Tracking using Voxel Data
Submitted to the International Journal of Computer Vision Human Body Model Acquisition and Tracking using Voxel Data Ivana Mikić 2, Mohan Trivedi 1, Edward Hunter 2, Pamela Cosman 1 1 Department of Electrical
More informationM 3 : Marker-free Model Reconstruction and Motion Tracking
M 3 : Marker-free Model Reconstruction and Motion Tracking from 3D Voxel Data Edilson de Aguiar, Christian Theobalt, Marcus Magnor, Holger Theisel, Hans-Peter Seidel MPI Informatik, Saarbrücken, Germany
More information6.801/866. Segmentation and Line Fitting. T. Darrell
6.801/866 Segmentation and Line Fitting T. Darrell Segmentation and Line Fitting Gestalt grouping Background subtraction K-Means Graph cuts Hough transform Iterative fitting (Next time: Probabilistic segmentation)
More informationGrasp Recognition using a 3D Articulated Model and Infrared Images
Grasp Recognition using a 3D Articulated Model and Infrared Images Koichi Ogawara Institute of Industrial Science, Univ. of Tokyo, Tokyo, Japan Jun Takamatsu Institute of Industrial Science, Univ. of Tokyo,
More informationGeneralized Principal Component Analysis CVPR 2007
Generalized Principal Component Analysis Tutorial @ CVPR 2007 Yi Ma ECE Department University of Illinois Urbana Champaign René Vidal Center for Imaging Science Institute for Computational Medicine Johns
More informationCurve Subdivision in SE(2)
Curve Subdivision in SE(2) Jan Hakenberg, ETH Zürich 2018-07-26 Figure: A point in the special Euclidean group SE(2) consists of a position in the plane and a heading. The figure shows two rounds of cubic
More informationDETC APPROXIMATE MOTION SYNTHESIS OF SPHERICAL KINEMATIC CHAINS
Proceedings of the ASME 2007 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference IDETC/CIE 2007 September 4-7, 2007, Las Vegas, Nevada, USA DETC2007-34372
More informationRevising Stereo Vision Maps in Particle Filter Based SLAM using Localisation Confidence and Sample History
Revising Stereo Vision Maps in Particle Filter Based SLAM using Localisation Confidence and Sample History Simon Thompson and Satoshi Kagami Digital Human Research Center National Institute of Advanced
More informationFitting (LMedS, RANSAC)
Fitting (LMedS, RANSAC) Thursday, 23/03/2017 Antonis Argyros e-mail: argyros@csd.uoc.gr LMedS and RANSAC What if we have very many outliers? 2 1 Least Median of Squares ri : Residuals Least Squares n 2
More informationLecture VI: Constraints and Controllers
Lecture VI: Constraints and Controllers Motion Constraints In practice, no rigid body is free to move around on its own. Movement is constrained: wheels on a chair human body parts trigger of a gun opening
More informationSupervised texture detection in images
Supervised texture detection in images Branislav Mičušík and Allan Hanbury Pattern Recognition and Image Processing Group, Institute of Computer Aided Automation, Vienna University of Technology Favoritenstraße
More informationArticulated Model Based People Tracking Using Motion Models
Articulated Model Based People Tracking Using Motion Models Huazhong Ning, Liang Wang, Weiming Hu and Tieniu Tan National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences,
More informationProbabilistic Observation Models for Tracking Based on Optical Flow
Probabilistic Observation Models for Tracking Based on Optical Flow Manuel J. Lucena 1,JoséM.Fuertes 1, Nicolas Perez de la Blanca 2, Antonio Garrido 2,andNicolás Ruiz 3 1 Departamento de Informatica,
More informationGEOMETRIC TOOLS FOR COMPUTER GRAPHICS
GEOMETRIC TOOLS FOR COMPUTER GRAPHICS PHILIP J. SCHNEIDER DAVID H. EBERLY MORGAN KAUFMANN PUBLISHERS A N I M P R I N T O F E L S E V I E R S C I E N C E A M S T E R D A M B O S T O N L O N D O N N E W
More informationCS545 Contents IX. Inverse Kinematics. Reading Assignment for Next Class. Analytical Methods Iterative (Differential) Methods
CS545 Contents IX Inverse Kinematics Analytical Methods Iterative (Differential) Methods Geometric and Analytical Jacobian Jacobian Transpose Method Pseudo-Inverse Pseudo-Inverse with Optimization Extended
More information